Reproduce “Physical Words: A Vocabulary for Structuring Latent Reasoning in Robot Manipulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiaming Liu, Xinghao Zhu, Yiming Zhang, Shanghang Zhang, Yinxi Wang, Yichu Yang, Zhuoyang Liu, Dantong Niu and 3 more
PhysWord helps robot policies anticipate task stages by predicting action words alongside future visual or tactile representations. Its 16-word vocabulary anchors temporally ordered predictions before action generation. Across eight visuotactile UniVTAC tasks, it averages 90.3% success, and across ten vision-only LIBERO-Long tasks, 99.6%. In a three-cycle paper task, it reaches 70.0%; its vocabulary excludes whole-body actions such as climbing stairs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.